How To Find Out How Much A Doctor'S Service Costs In Massachusetts

Transcription

1 FEDERAL RESERVE BANK OF SAN FRANCISCO WORKING PAPER SERIES Physician Payments Under Health Care Reform Abe Dunn Adam Hale Shapiro, Federal Reserve Bank of San Francisco March 2014 Working Paper The views in this paper are solely the responsibility of the authors and should not be interpreted as reflecting the views of the Federal Reserve Bank of San Francisco or the Board of Governors of the Federal Reserve System.

2 Physician Payments Under Health Care Reform Abe Dunn and Adam Hale Shapiro March 28, 2014 Abstract This study examines the impact of major health insurance reform on payments made in the health care sector. We study the prices of services paid to physicians in the privately insured market during the Massachusetts health care reform. The reform increased the number of insured individuals as well as introduced an online marketplace where insurers compete. We estimate that, over the reform period, physician payments increased at least 10.8 percentage points relative to control areas. Payment increases began around the time legislation passed the House and Senate the period in which their was a high probability of the bill eventually becoming law. This result is consistent with fixed-duration payment contracts being negotiated in anticipation of future demand and competition. We thank Ana Aizcorbe, Jeff Clemens, Randy Ellis, Bart Hobijn, Amanda Kowalski, Nancy Turnball, Rob Valletta and seminar participants at the Federal Reserve Bank of San Francisco for comments and suggestions. We also thank Erika Wilkinson of Blue Cross Blue Shield of Massachusetts and David McGuire of Partners Healthcare for helpful discussions. The views expressed in the paper are those of the authors and not of the Federal Reserve System or the Federal Reserve Bank of San Francisco Corresponding Author: Federal Reserve Bank of San Francisco, [email protected]

3 1 Introduction The primary goals of the 2006 Massachusetts health care reform were to expand the number of individuals with health insurance coverage and increase the degree of competition in the insurance marketplace. The expansion focused on using financial incentives to spur enrollment in the private insurance market. The key elements of the legislation included an individual insurance mandate, employer requirements to provide insurance, the creation of a subsidized insurance program to low-income individuals, and the introduction of a health insurance exchange. The reform has been quite successful in its goal, with the percentage of individuals without insurance falling from 6.4 in June of 2006 to 1.9 in 2010 (Massachusetts Division of Health Care Finance and Policy (DHCFP), 2011). The reform in Massachusetts served as a model for the 2010 national health-care reform legislation, known as the Affordable Care Act (ACA). Both the Congressional Budget Office (CBO) and the Center for Medicare and Medicaid Services (CMS) project that over 30 million more individuals will have insurance in the United States over the next decade as a result of the reform. 1 Examining the case of Massachusetts may therefore give important insights about the impact of national reform. Recent studies have looked at the causal impact of the Massachusetts reform on a wide array of variables including healthcare demand (Kolstad and Kowalski [2012] and Long, Stockley, Dahlen [2012]) and the place of service (i.e., emergency room or physician office) (Miller [2012a]). The finding that the reform affected service utilization is in line with a vast empirical literature that documents the effect of insurance coverage on utilization (e.g., Manning et al. [1987], Finkelstein [2007], and Finkelstein et al. [2012]). However, distinct from prior expansions in public insurance, where prices are fixed by regulators(i.e., Medicare and Medicaid), both the ACA and the reforms in Massachusetts rely more heavily on the private sector. Consequently, these reforms may not only affect utilization, but may also have an impact on equilibrium prices, including insurance premiums and payments to health care providers. This paper focuses specifically on the effect of insurance expansion on payments to providers. Since health care accounts for about 18 percent of nominal GDP, even a relatively small price increase to providers would increase the nominal expenditures devoted to the health sector and also affect national measures of inflation and output. Prices are also important signals that affect the long-run market entry decisions and also short-run decisions on the quantity and types of services offered to patients. 2 Theoretically, the 1 Survey evidence from Krueger and Kuziemko [2013] suggest that 35 million uninsured individuals would gain insurance. 2 Clemens and Gottlieb [2013] find supply motives associated with physician payment increases, indi- 2

4 mechanisms by which the reform might impact provider fees are straightforward. First, the reform led more than 400,000 previously uninsured individuals to obtain health insurance, causing a substantial increase in the demand for health care (Kolstad and Kowalski [2012], Miller [2012b], and Long, Stockley, Dahlen[2012]). To match demand, insurers may need to adjust physician payments in order to maintain their current physician network or to draw new physicians into their network. Second, the new health insurance exchange, known as the Massachusetts Connector, likely increased the degree of competition among insurers (Ericson and Starc [2012]). Dunn and Shapiro [2013] and Dafny et al [2012] show that a higher degree of competition in the insurance market raises payments to physicians. While theory would suggest that expansion would likely lead to higher prices, the magnitude and timing of the effect is unclear. For instance, there is the possibility that physician capacity was sufficient to accommodate expansion at existing rates. Ultimately, the effect of the reform on prices must be measured empirically. In this study, we focus on the impact of the Massachusetts reform on payments to physicians. There are two reasons for focusing on physician prices. First, relative to how other payments are set in the healthcare industry, physician payment contracts are usually quite simple. Prices are based on fees for a specific procedures defined by Current Procedural Terminology (CPT) codes. For example, there are distinct CPT codes for office visits that last 15 minutes and those that last 30 minutes. 3 This is in contrast to how prices are negotiated for inpatient hospital services where the definition of a service is more complex and the pricing methodology may change dramatically depending on the specific contract between each insurer and provider (see Reinhardt [2006]). 4 A second reason for focusing on physician prices is that the number of physicians is quite large reducing the chance that the price changes are caused by a single provider s negotiating practices. Prior studies measuring the impact of Massachusetts reform have relied on two different sources of variation to conduct difference-in-difference analysis: across-state variation and across-county variation. We apply both approaches, obtaining two distinct estimates of the effect on physician prices. Specifically, one approach uses across-state variation to identify the relative change in Massachusetts prices around the time of the reform, relative to cating the payments may also affect the quantity and types of services provided to patients. 3 Due to the thousands of CPT codes in existence, physicians usually negotiate the prices of all CPT codes at once based on a fee schedule for example, all prices set relative to Medicare prices (see Clemens and Gottlieb [2014]). 4 For example, inpatient hospital contracts may be based on a discount off of charges, a per diem rate, or price based on the diagnosis code (i.e., DRG code) of the patient. 3

5 comparable states, as applied in Kolstad and Kowalski [2012]. To obtain comparable control states, we apply the synthetic control approach of Abadie, Diamond, and Hainmueller [2010]. 5 The second approach exploits variation in the pre-reform uninsured rates in the county, as applied by Miller [2012a]. 6 This analysis adds another layer of variation the uninsured rate of the county under the presumption that providers and insurers residing in those counties with higher uninsured rates should have anticipated a higher impact of the reform on insurance coverage than those counties with already near-full coverage. Unlike prior studies of the reform, our analysis assesses the timing of the impact of the reform. In examining price impacts of the reform, tracking the timing is important since key industry characteristics suggest that a response to the reform should be expected prior to implementation. First, price negotiations between physician firms and insurers take place rather sporadically anywhere from annually to every five years. Since prices are essentially stuck for a fixed duration of time, it is likely that insurers set prices based on expectations of future demand and competition over the contract period. Strategically, insurers need to negotiate contracts with a enough time remaining to advertise their provider networks to consumers. 7 Additional dynamic considerations arise from the large switching costs in health insurance markets that lock individuals into a plan (Ericson [2011], Handel [2011], and Nosal [2012]). In industries with high switching costs, firms that do not respond immediately to expected changes in market conditions risk substantial profit loss in future periods (Klemperer [1995] and Farrell and Klemperer [2007]). 8 Both our across-state and across-county analysis lead to a similar conclusion. Relative to control states and counties, we find that prices are significantly higher post-reform, relative to the pre-reform period. Overall, our estimates imply that at least one-sixth of overall physician service price growth in Massachusetts was directly attributable to the reform itself. The timing of the price increase occurs around the time the health care legislation passed the house and senate the period in which their was a high probability 5 Our state-level analysis is based on a variation of the classical Fisher permutation test similar to a recent paper by Buchmueller, DiNardo, and Valletta [2012]. Similar results are found using alternative difference-in-difference approaches at the state level. 6 This approach is also applied in Finkelstein [2007] looking at the effects from the introduction of Medicare. 7 The first major components of the legislation aimed at people with lower incomes went into effect in October 1, 2006 and January 1, In other words, anticipated future changes in market conditions will have immediate effects on prices in markets with switching costs Klemperer [1995]. The expansion of insurance to the uninsured likely provided a key opportunity for insurers with low market share to expand, since uninsured individuals arguably have no switching costs. 4

6 of the bill being signed. This timing is consistent with recent microeconometric studies in other industries that show price changes in response to anticipated changes in the competitive environment. For example, Goolsbee and Syverson [2008] show that rival airlines set prices in anticipation of Southwest Airline s entry in a market; and Tenn and Wendling [2013] show that generic drug manufacturers reduce prices in anticipation of potential entry by other generic firms. 9 It is important to highlight a few caveats of this study. First, this study assesses the impact of the reform on payments made to physicians from insurers, not the impact of the reform on insurance premiums or out-of-pocket costs that are charged to employers and enrollees. Given the reform induced expansion in the private insurance market, ultimately changing the competitive landscape, it is possible that no medical-care price increases were passed onto consumers. 10 In fact, Graves and Gruber [2012] find that the reforms had no effect on group premiums in Massachusetts relative to other states, and led to a reduction in premiums in the non-group market. Second, our study assess the prices of services provided by physicians in an office setting. We make no definitive assessment about pharmaceutical prices or payments made to hospitals in an inpatient setting. Third, we analyze the prices of physicians services in the privately insured market. Our data does not include Medicare or Medicaid prices or prices charged to the uninsured. Furthermore, we likely do not see prices paid by plans in the subsidized insurance exchange. The overall impact of the reform would be partly offset if subsidized plans pay physicians less generously than other plans, or previously uninsured patients pay lower prices after the reform. Finally, the results in this study should not be treated as a normative statement regarding the merits of the reform. To fully evaluate the success of a reform, policymakers must weigh the potential benefits of the reform, such as expanded coverage, improved health outcomes, and a more competitive insurance market, against the associated cost. 9 Kwoka and Shumilkina [2010] show that prices fall in airline markets when potential competitors are eliminated after a merger. Anticipatory effects are not confined to prices. For instance, Goetz and Shapiro [2012] find that airlines preemptively codeshare when a route is threatened, while Ellison and Ellison [2011] find that pharmaceutical firms make strategic investments to deter future entry. It is interesting to note that forward-looking price setting models have been commonplace in the macroeconomics literature for some time (e.g. see Gali and Gertler [1999]). 10 Graves and Gruber [2012] find that the reforms had no effect on average premiums in Massachusetts relative to other states. DHCFP [2011] reports profit margins for insurers falling post reform. Dafny et al. [2012] shows a strong relationship between premiums and insurer concentration. A full accounting of consumer welfare would also need to consider the gains for those that receive insurance who previously had none. 5

7 2 Massachusetts Health-Care Reform 2.1 Timeline of Events Health-care reform efforts in Massachusetts began in early 2003 with the inauguration of Mitt Romney as governor. Upon taking office, Romney s staff immediately began consulting with policy experts and academics to figure out ways to cover the uninsured. By May 2004, a preliminary insurance coverage plan was created by Romney s health and human services secretary, who privately circulated a 29-page white paper outlining a plan to increase coverage by expanding the Medicaid program and imposing individual and employer mandates. In hindsight, the plan described in this white paper looked very similar to the actual legislation that was eventually enacted two years later. 11 Romney did not fully advertise his ideas concerning health insurance reform to the general public until November 2004, when he wrote an editorial in The Boston Globe outlining his ideas concerning a health insurance overhaul. The editorial explicitly came out against an insurance mandate. 12 Romney instead focused on how the state could lower health care costs via offering health insurance with larger deductibles and smaller benefits. 13 Remarkably, seven months later in June 2005, Romney publicly changed his strong views regarding an insurance mandate. Specifically, in a keynote address at an event organized by the Blue Cross Blue Shield of Massachusetts Foundation, he discussed that his plan would require all residents to have some form of health insurance or agree to pay their health care expenditures out of pocket. 14 Romney touted the individual mandate as the the ultimate conservative idea. 11 The paper was released to the public in January Thesecondparagraphoftheeditorialread, NextyearIamcommitted toworkingwiththe Legislature to pass a comprehensive, market-based reform program for healthcare. It will not be a governmentmandated universal coverage pay or play scheme nor a single payer system. It will not require new taxes. What it will do is restrain the growth in healthcare costs and change how we provide healthcare for those who receive it at taxpayer expense. And, it can lead to every citizen in Massachusetts having health coverage. I call it Commonwealth Care. 13 Romney writes later in the editorial: Insurers tell us they can develop plans costing less than half of today s standard rate of $500 for an individual. These plans still provide primary, preventative, specialty, and catastrophic care. The cost could be lower with higher deductibles and more restrictions. New York introduced a programin which private insurers offer rates as low as $140a month. We can have a similarly affordable program in Massachusetts: Commonwealth Care Basic. 14 See The Boston Globe June 22, 2005 article by Scott Greenberger local/articles/2005/06/22/romney_eyes_penalties_for_those_lacking_insurance 6

8 The next important milestone occurred later that year, in November 2005, when the Massachusetts House and Senate each passed separate health insurance reform bills. Both bills included an individual mandate, although the Senate bill was less comprehensive, in that it did not include expanded subsidized coverage or an employer assessment (Bebinger [2012]). It took a few months of negotiations between both sides of the legislature on an employer assessment until a compromise was finally reached in April The Massachusetts legislation made national headlines and eventually served as a benchmark for national health care reform under the ACA of The Three-Legged Stool Design Similar to the ACA of 2010, the Massachusetts 2006 legislation was based on a threelegged stool design (Gruber [2011]). The first leg included private insurance market reforms. These aimed to limit price discriminating against individuals with pre-existing conditions, guarantee issuance and renewal of insurance, and prohibit medical underwriting (McDonough et al. [2008]). To increase the availability and transparency of health insurance, an exchange was created, called the Commonwealth Health Insurance Connector. The second leg of the stool aimed to keep the market from unraveling. For instance, with guaranteed issuance and limits on pre-existing conditions, healthy individuals may find it in their interest to wait to obtain health insurance until they finally need health care. For this reason, part of the reform consisted of an insurance mandate, which required individuals over the age of 18 to purchase insurance or pay a penalty. The third leg of the stool represented government subsidies to allow low-income individuals access to the insurance market. Specifically, Massachusetts created the Commonwealth Care Health Insurance Program (CommCare), which would be sold on the Massachusetts Connector. This program gave eligible people with incomes below 150 percent of the poverty level insurance coverage without any premium. Eligible households with incomes between 150 and 300 percent of the poverty line paid premiums for CommCare based on a gradient. It is important to note that the Massachusetts Connector also served as a marketplace for nonsubsidized plans known as Commonwealth Choice plans. These plans are designed for uninsured individuals who are not eligible for Medicaid or CommCare and must meet certain standards for quality and value before given a seal of approval. The reform also included a number of other ordinances. For example, to expand coverage, the legislation encouraged employers to offer insurance. This was done by imposing a penalty 7

9 ($295 per worker) to firms with eleven or more full-time equivalent employees who do not make fair and reasonable contributions to employee health insurance costs. 2.3 Impact of the Reform Estimates vary on the exact effect of the reform on coverage, although the consensus is that the impact was quite large. The Massachusetts DHCFP estimated that insurance enrollment increased by 411,722 individuals between June 2006 and 2010 a 4.5 percentage point increase in coverage. Other studies find that the increase in coverage was considerably larger. For instance, Kolstad and Kowalski [2012] perform a difference-in-differences estimation and find that insurance coverage increased by approximately 6 percent due to the reform. Long, Stockley and Yemane [2009] perform a similar difference-in-differences estimation and find a 6.6 percentage point increase in coverage. 15 Miller [2012a] shows that the insurance coverage increase mainly occurred in 2007, while Kolstad and Kowalski [2012] find that it occurred in Coverage expansion from the reform was a result of both an increase in coverage from public insurance programs and employer-sponsored insurance (ESI). Both Kolstad and Kowalski [2012] and Long, Stockley, and Yemane [2009] find that approximately half of the increase in insurance coverage was a take-up of employersponsored insurance. 16 Afewrecentstudieshavefoundthattheincreaseininsurancecoverageledtoanincrease in health-care demand. Long, Stockley, and Dahlen [2012] find that the overall percentage of people who visited a general doctor in the last 12 months increased by 2.1 percentage points between 2006 and 2010, and the percent who visited a specialist increased even more, 3.7 percentage points. 17 Miller [2012b] also finds a significant increase in doctors 15 Long, Stockley, and Dahlen [2012] look at the Massachusetts Health Reform Survey (MHRS) and find that the insurance coverage rate increased from 86.6 percent in 2006 to 94.2 percent in 2010 an increase of 7.6 percentage points. 16 The DHCFP reports that the percentage of Massachusetts employers offering ESI increased from 70 percent in 2005 to 77 percent in Long, Stockley,and Dahlen [2012] find that the percent of residents with ESI increased from 64.4 percent in 2006 to 68.0 percent in In 2010, approximately 91 percent of employees worked for a firm that sponsored insurance. 17 The evidence thus far indicates that the reform has not had an effect on access to care. Gruber [2011], for instance, cites a survey from the Massachusetts Medical Society that found average wait times remained constant over the course of the reform. Long and Stockley [2010], who examine the Massachusetts Health Reform Survey, find that barriers to getting care, either because the physician was not accepting new patients or not accepting their health insurance, remained high throughout the reform approximately 20 percent of patients over the reform period had trouble finding a doctor. Long [2010] explains that problems with individuals finding the care they need were exacerbated in the years immediately following 8

10 visits. The increase in the demand for services does not appear to have led to movements of physicians into the Massachusetts market. In Table 1 we report the number and geographic movement of doctors in the United States from the SK&A database. The database spans 2005 to 2008 and includes information on the name, location, and specialty of physicians (see Dunn and Shapiro [2013] for more information about the SK&A database). 18 Table 1 provides counts for those physicians who were in the database all four years, which allows us to measure the geographic movement of physicians who practiced in Massachusetts before the reform took place. 19 The numbers show that there was little movement of physicians into or out of Massachusetts during the reform. 20 We also consulted with Partners Healthcare, a large health system in Massachusetts, who informed us that although provider consolidation has increased in recent years, there was not a lot of consolidation occurring around the reform period. Table 1: Physician Geographic Movement Before, During, and After Reform 2005h1 2005h2 2006h1 2006h2 2007h1 2007h2 2008h1 2008h2 CT 2,510 2,505 2,507 2,508 2,505 2,504 2,503 2,501 MA 4,288 4,299 4,297 4,293 4,300 4,305 4,303 4,297 ME NH RI VT OTHER 151, , , , , , , ,634 Total 161, , , , , , , ,372 Notes: This table reports the number of physicians in the SK&A physician database during the period. Physicians who entered or left the data mid-sample are not reported. The sample includes internists, family doctors, general practitioners, pediatricians, cardiologists, orthopedists, endocrinologists, gynecologists, gastroenterologists, and vascular surgeons. Overall, health-care expenditures and premiums grew during the 2000s in Massachusetts. The DHCFP reports that private spending per member grew 15.5 percent between 2006 the reform and then eventually improved. 18 As noted in our previous work, this data is very noisy concerning consolidation in a time-series dimension. However, it is comprehensive in terms of individual physician counts. 19 This also allows us to control for the fact that the size of the SK&A sample grew over their sample period. 20 We also observe relatively little movement in the share of doctors across states when looking at county data from the American Medical Association. 9

11 and 2008, with physician services accounting for approximately 40 percent of this growth. Spending per capita on physician services in the privately insured market grew 8.4 percent between 2006 and 2007 and then 9.8 percent between 2007 and Spending on physician services then increased 11.8 percent between 2008 and Most spending growth was attributed to growth in service prices. Premiums also grew, although not as rapidly as medical-care spending, implying that insurance carriers medical loss ratios were rising. Premiums grew 5 to 10 percent annually between 2007 and 2009, while the medical loss ratio increased from 88 percent to 91 percent. 2.4 Physician Service Prices and the Reform Prices for physician services in private insurance markets are set through negotiations between insurers and physician groups. Both sides may gain from contracting and agreeing on rates before marketing insurance plans. Those physicians who agree to a contract are placed in-network, where enrollees may visit a physician at a lower out-of-pocket cost than out-of-network physicians. Physicians gain access to enrollees in the insurer s plan, while the insurer is more likely to attract enrollees if the network contains a broad, high-quality physician network. Each side may attempt to gain leverage and more favorable rates by threatening to not contract with the other party. Contract durations vary by the health care provider and the insurance firm. For instance, contracts between large health systems and commercial payers are typically negotiated every three years, sometimes up to every five years, whereas contracts between smaller practices and insurers are typically set on an auto-renewing annual basis with provisions that allow either party to terminate the contract. There are at least two reasons why service prices could be affected by the Massachusetts reform. The first reason is related to insurance demand. The individual and employer mandate increased the number of insured by over 400,000 individuals. To match the increased demand, an insurer may be enticed to draw new physicians into its network with higher fees. This effect could potentially be larger in areas with a high number of uninsured individuals. The second reason is related to the competitive environment in the insurance market. All else equal, more competition between insurers acts to raise physician payments (Dunn and Shapiro [2013]). The expanded enrollment and the exchange may have acted to increase competition by bringing in new entrants from the managed Medicaid market 21 as 21 Plans such as Network Health, Neighborhood Health Plan, and Healthnet, entered the non-medicaid 10

12 well as by lowering consumer search costs. 22 Besides competing on premiums, insurers compete in terms of the size of their physician networks. For instance, the Massachusetts Connector allows the shopper to search for plans that include a specific physician. All else equal, higher payments encourage physicians to stay in(or join) a network which maintains (or increases) the insurer s network size. These competitive effects might be especially true in areas of Massachusetts where consumers are more apt to use the Connector. It is important keep in mind the possibility that the reform had no effect on service prices. In particular, physicians may have sufficient capacity or may be able to adjust practice patterns, allowing them to see more patients at pre-reform rates. The ambiguity of the theoretical prediction elevates the importance of testing the effect empirically. 3 MarketScan R Data The MarketScan R database used in this analysis tracks insurance claims from physicians using a nationwide convenience sample of patients. 23 The data span 2003 through 2010 and include payments in an office setting to cardiologists, orthopedists, internists, gastroenterologists, gynecologists, endocrinologists, and family practice physicians. The data include health claims from employers and insurance carriers throughout the entire United States; all claims have been paid and adjudicated. Geographic information is provided about the patient s county of residence and physician s county of practice. Although the MarketScan R database is a convenience sample, Dunn, Shapiro, and Liebman [2012] show that data are actually quite representative of actual health-care spending. Each observation in the data corresponds to a service line item in an explanation of benefits form. We use MarketScan s payment variable, which is defined as the total gross payment to a provider for a specific service. Each service line represents the Current Procedural Terminology(CPT) code including any modifiers performed on the patient. For instance, the most common CPT code is 99213, which represents a 15-minute establishedmarket by going on the exchange to serve individuals required to get commercial insurance or individuals obtaining subsidized insurance 22 The Connector allows consumers to compare the prices and benefits of different plans on a single web page. Studies have shown that, more generally, the internet increases competitiveness. For example, Brown and Goolsbee [2002] find that in the life insurance industry, an increase in the share of individuals using the Internet reduces insurance premiums. Dafny et al. [2012] and Dunn and Shapiro [2013] show that decreased insurer market power increases physician income and payments. 23 This paper uses the Commercial Claims and Encounters Database portion of the MarketScan R Databases, which includes records at the encounter level. 11

13 Table 2: MarketScan R Sample U.S. (excl. MA) New England (excl. MA) Massachusetts Number of Procedures Number of Provider IDs 98,988 3,988 1,757 Observations 209,212,900 3,311,833 1,741,223 Average Patient Age Percent Spending by Plan Type PPO 62.7% 42.9% 20.1% HMO 13.2% 17.0% 51.8% POS 11.5% 27.0% 19.4% Other 12.6% 13.1% 8.7% Percent Spending by Procedure 15 min est. E/M (99213) 32.7% 33.1% 30.7% 25 min est. E/M (99214) 23.8% 20.0% 24.1% Comprehensive E/M (99396) 6.6% 14.8% 10.9% 30 min new E/M (99203) 4.0% 3.0% 2.4% 40 min est. E/M (99215) 3.9% 1.7% 3.3% Remaining 29.0% 27.4 % 28.6% patient office visit. The payment of each service line represents the amount of dollars eligible for payment after applying pricing guidelines such as fee schedules and discounts, and before applying deductibles and copayments. For our analysis, we kept only those CPT codes where there existed at least 100 observations per quarter in Massachusetts. This left 33 CPT codes, which make up 66 percent of all physician spending in MarketScan R. Typically, providers negotiate all procedures from a percentage off of a base fee schedule, for example, 20 percent above Medicare rates. 24 A summary of the MarketScan R sample used in this study is shown in Table 2. MarketScan R includes a variable called provider ID, which is an encrypted identification number of the health-care provider. This variable may represent a single physician, a group of physicians, or the hospital employing the physician. The important information for our study is that this variable keeps track of a given health-care provider over time. This allows us to control for new health-care providers spuriously entering or leaving the MarketScan R sample. 25 To keep the sample balanced in this regard, we keep only those 24 See Clemens and Gottlieb [2013] 25 Since this variable is created by clients of Truven Health, there may be a single physician or physician firm (i.e. group) with multiple provider IDs, if the same physician is in the data provided by multiple sets of clients. The ID variable should therefore not be used to track statistics on the number of physicians in 12

14 provider IDs that are in the data at least six consecutive years. This step removed only 3 percent of the observations in the data, leaving about 100,000 provider IDs in the United States and 1,757 in Massachusetts. We show that the results are robust to a range of provider ID cutoffs. MarketScan R alsoindicatesthetypeofinsuranceplantheclaimwasmadeunder, which allowsustoignoreepisodesinwhichacapitationpaymentwasmade. 26 Thebottomportion of Table 2 compares the types of spending in Massachusetts to the entire United States and New England. 27 An apparent disparity in Massachusetts is its low share of PPO spending and high share of HMO spending. We verified this with HealthLeaders-InterStudy (HLIS), which shows similar HMO market shares and verifies that this is likely representative. 28 Nevertheless, speaking with insurance companies in Massachusetts leads us to believe that the distinction between plan types has blurred over the past couple of decades. That is, there are many HMO plans with low restrictions and PPO plans with high restrictions. The cutoff in the distinction between these types of plans likely varies regionally. 4 State-Level Analysis Our state-level analysis examines overall physician service price growth in Massachusetts over certain periods of time during the course of the health care reform. For instance, a simple difference-in-differences estimate would compare Massachusetts price growth before and after the reform relative to the difference in growth rates of non-massachusetts states before and after the reform. We take a more flexible approach by breaking the sample into three period periods. We characterize a lead-up period, which begins in 2004q4, the quarter in which Governor Romney wrote the The Boston Globe editorial, and extends to the quarter the reform was implemented, 2006q2. The two periods around the lead-up a market. 26 Approximately 3 percent of our sample are capitated episodes. These observationsare likely to include closed HMO systems such as Kaiser-Permanente patients. 27 New England states include Connecticut, Maine, Massachusetts, New Hampshire, Rhode Island, and Vermont. Results did not change when we excluded Vermont from our sample. Vermont passed the Health Care Affordability Acts in 2006, which aimed to increase insurance enrollment, but did not include an insurance mandate nor a penalty to firms for not providing employer sponsored insurance. See Deprez et al. (2010). 28 HLIS reports an average HMO market penetration of 49.3 percent in Massachusetts and 27.9 percent in the United States from 2004 through PPO market penetration was 36 percent in Massachusetts and 61.3 percent in the United States. In our main empirical specification, we include the type of plan in the fixed effect. Results were quantitatively similar without plan type included. 13

15 period we simply refer to as the pre- and post-reform periods. 29 As noted by Moulton [1990], statistical inference is not straightforward in this setting. Due to the focus on a single state, the standard asymptotic assumptions under a clustering framework do not hold. To address this issue we use a variation of the Fisher permutation test, similar to that performed in Buchmueller, Dinardo, and Valletta [2011]. 4.1 Service Price Growth by State Our exercise begins by obtaining price-growth estimates by running a simple fixedeffects regression for each state plus the District of Columbia. Each regression tracks price growthattheservicelevel, wherewedefinetheserviceleveltobeaspecificcptcode, plantype (e.g. HMO, PPO, etc.), and physician ID triple. We regress the logarithm of service price on time-period dummies (by quarter) as well as time-period dummies interacted with a state dummy. That is, for each state s, we run: t=2010:q4 ln(p jct ) = t 1(STATE = s) 1(Time t )+δ t 1(Time t t=2003:q2[γ )]+ [α jc 1(Service jc )]+ε jnct jc (1) where j indicates a CPT code-physician ID-plan triple, and c indicates a county. 30 Growth rates can then be calculated from the γ t estimates. 31 We can then treat the other 50 estimates of price growth as the sampling distribution for general price growth over any two periods of time t 1 and t 2. Accordingly, the test statistic for the null hypothesis that Massachusetts price growth during a specific period of time is no different than that any other state can be obtained by computing the percentile that it falls in the entire distribution. Since there are 50 placebo estimates, if Massachusetts is ranked second from the top or bottom of the distribution, we can infer that its price growth is statistically significant at the 10 percent level. Statistical significance at the 5 percent level requires it be ranked first. Each panel in Figure 1 consists of a histogram of price-growth rates of the 51 states. Note that our regression estimate γ t measures price growth for each state relative to the national average for instance, a price growth rate of 5 percent means that prices grew 5 29 Weconductamoresensitivetestofthespecifictimingofthereformwhenweconducttheacross-county analysis. 30 The regressions are weighted by the average price of the CPT code over the entire sample as would be done in a Tornqvist index. This gives less weight to lower value services. 31 For example, the cumulative growth between 2003q1 and 2010q4 is exp(γ 2010q4 ) 1. 14

16 percentage points faster than the U.S. average. 32 The top-left panel depicts the cumulative growth rate between 2003q1 and 2010q4. The other three panels report annualized growth rates during the three specific time periods outlined above. Figure 1: Physician Service Price Growth by State Total Price Growth: Full Sample (2003q1 to 2010q4) Annualized Price Growth: Pre-Reform (2003q1 to 2004q4) Frequency Frequency MA 6 4 MA Annualized Price Growth: Lead-up to Reform (2004q4 to 2006q2) Annualized Price Growth: Post-Reform (2006q2 to 2010q4) Frequency Frequency MA 2 1 MA 0 0 Notes: Histogram of physician service price growth by state relative to the U.S. average price growth. Massachusetts value in the histogram is marked. The histograms are not centered at zero growth because prices in some large states (e.g. Texas) grew well below average. The top left panel shows that physician prices in Massachusetts grew more than in any other state over the 2003q1 to 2010q4 period representing significance at the 5 percent level. Prices in Massachusetts grew 49 percentage points faster than overall U.S. prices duringthisperiod. 33 Forcomparison, thestateswiththenexthighest pricegrowtharenew Hampshire (37 percent), North Dakota (36 percent), and Oregon (28 percent). Relative to the five other New England states, prices in Massachusetts grew 28 percentage points 32 Note that the histogram is not centered at zero growth because prices in some large states (e.g. Texas) grew well below average. 33 This is a total of 69 percent growth in payments over the 2003 to 2010 period 15

17 faster. The other three panels show that, on an annualized basis, Massachusetts is in the upper tail of the distribution in terms of price growthduring all three time periods. However, it is only in the lead-up period that Massachusetts had the highest overall growth. Specifically, relative to average U.S. growth, Massachusetts prices grew 5 percent per year during the pre-reform period, 6.4 percent per year during the lead-up period, and 3.7 percent per year in the post-reform period Synthetic-Control Analysis As a second exercise, we perform an extension of the above analysis by applying the synthetic control approach of Abadie, Diamond, and Hainmueller [2010]. As explained by these authors, a synthetic control state can be constructed as a weighted average of the available control units the other 49 states and the District of Columbia. Applying this technique, it follows that a synthetic measure of γ t can be constructed for Massachusetts, γ synth t the price level of a synthetic state that resembles Massachusetts prior to the reform. To do so requires using a set of predictor variables. Our predictor variables are the percent of the population over age 65, the uninsured rate, median household income, median house value, median rent, percent of population with a college degree, population density, the unemployment rate, percent of spending that is HMO (from MarketScan), percent of spending that is PPO (from MarketScan), the number of university hospitals per capita, and the pre-reform estimates of γ t. 35 Means of these predictor variables and state weights for Massachusetts are available in the appendix. The synthetic control for Massachusetts was determined to be a convex combination of Rhode Island, New Hampshire, Illinois and Alaska. Results were generally robust to a range of predictor variables. 36 Our empirical exercise is performed as follows. For each quarter, t, we construct a measure of the gap between the synthetic control price level and the actual price level: γ gap t = γ t γ synth t. For the case of Massachusetts, this creates a series, γ gap t, which is designed to track how its price level differs from what would have happened if health care reform had not happened. By looking at the change in γ gap t between any two time periods, 34 As Kolstad and Kowalski [2012] refer to 2006q2 to 2007q2 as the implementation period, we also examined a post-implementation period, 2007q2 to 2010q4. The growth rates in the post-implementation period looked very similar to the overall post-reform period. 35 We use γ 2004:1, γ 2005:1, and γ 2006:1. 36 As a robustness exercise we created synthetic controls without including any γ t as a predictor variable. These results are available in the appendix. 16

18 we can see how Massachusetts prices grew relative to its synthetic control over that time period. For statistical inference, we create a distribution of growth rates of γ gap t in which to compare the Massachusetts value of γ gap t. We do so by repeating this synthetic control exercise once for each of the other 50 states. That is for each state, we create its own synthetic control, and subsequently its own measure of γ gap t. In total, we create 51 series of γ gap t one series for Massachusetts, and 50 placebo series. This allows us to form a distribution of relative growths in the variable γ gap t between any two periods t 1 and t 2. Figure 5 displays the results of this exercise. Similar to what was suggested in Figure 1, the analysis using the synthetic control shows statistically significant price growth during the lead-up period. Massachusetts lies at the top of the distribution in terms of growth over the entire sample period, however, in terms of specific time periods, it is only during the lead-up period that Massachusetts price growth was considerably larger than its synthetic control. During the lead-up period, Massachusetts prices grew 4.3 percentage points (on annualized basis) faster than its synthetic control. The results of this exercise imply that the reform had a total effect on prices as large as 21 percent. A large portion of which stemmed from price growth occurring in the lead-up period. The results in the section are consistent with a story where services prices were negotiated up in anticipation of health care reform. Indeed, during the lead-up period there was extensive news coverage and political speeches about the reform. As price negotiations are based on lengthy contracts, anywhere from one to five years in length, insurers were likely taking this into account when setting fee schedules. As with any reduced-form analysis, there are some drawbacks to this analysis. Notably, identification of the difference-in-differences estimator at the state level rests mainly on a comparison in growth rates between time periods. To strengthen our identification in this regard, we next exploit variation within Massachusetts at the county level. Specifically, we run a triple difference-in-differences estimator by exploiting the fact that the reform should have had a stronger impact in those counties with higher rates of uninsured. 5 County-Level Analysis In our county-level analysis, we estimate whether the impact of the Massachusetts insurance reform was caused by the increase, or expected increase, in insurance coverage the reform would generate. To do so, we implement an identification technique similar to Finkelstein [2007] and Miller [2012]. Both Finkelstein and Miller exploited the fact that certain geographic areas should, ex ante, be more affected by policy reform than others 17

19 Figure 2: Physician Service Price Growth by State: Relative to Synthetic Control 16 Total Price Growth: Full Sample (2003q1 to 2010q4) 20 Annualized Price Growth: Pre-Reform (2003q1 to 2004q4) MA Frequency 8 6 Frequency MA Annualized Price Growth: Lead-up to Reform (2004q4 to 2006q2) 16 Annualized Price Growth: Post-Reform (2006q2 to 2010q4) Frequency 8 6 Frequency MA 2 MA 0 0 Notes: Histogram of physician service price growth by state relative to each state s synthetic control. For example, the cumulative growth between 2003q1 and 2010q4 is exp(γ gap 2010q4 ) 1, while annualized growth in the lead-up period is exp(γ gap 2006q2 γgap 2004q4 ) Massachusetts value in the histogram is marked. based on their pre-reform insurance coverage rates. In the setting of the Massachusetts reform, counties in Massachusetts with low rates of insurance coverage have the most coverage to gain by an insurance mandate. Health care providers and insurers located in low insurance coverage counties should therefore expect to be impacted more by the health insurance reform than those counties that already had high insurance coverage. 37 For instance, individuals who use the Connector are more apt to live in counties with high uninsured rates. Table 3 shows insurance coverage statistics for the non-elderly, taken from the U.S. Census Small Area Health Insurance Estimates (SAHIE) for the largest nine counties 37 Dunn and Shapiro [2013] show that the physician market may be as small as a 20-minute driving radius. 18

20 in Massachusetts. 38 The 2005 uninsured rate level varied among these counties from 9.5 percent in Norfolk and Worcester counties to 15.4 and 12.2 percent in Suffolk and Middlesex counties, respectively. As explained by Miller [2012], those counties with higher uninsured rates before the reform was implemented experienced proportionally larger increases in coverage during the reform. For example, Suffolk County experienced an 8.4 percent increase in insurance coverage while Worcester County experienced a 4.9 percent increase. Overall, the reform caused the uninsured rate to fall between 50 and 75 percent from its 2005 level. In other words, a Massachusetts county with a 1 percent higher uninsured rate than another Massachusetts county could rationally expect to see between a 0.50 to 0.75 percentage point larger increase in its insurance coverage rate. This leads to an exogenous causal interpretation because the reform had an expected differential impact across markets depending on the pre-reform size of the uninsured population. Table 3: County Statistics Suffolk Middlesex Norfolk Worcester Essex Plymouth Bristol Hampden Berkshire 2005 Uninsured Rate Coverage Rate 2005 to Percent of MA Spending in MarketScan Notes: Coverage Rate represents the percent increase in insurance coverage between 2005 and 2009 taken from the U.S. Census Small Area Health Insurance Estimates (SAHIE). Not included in this table are Dukes, Nantucket, Franklin, Hampshire, and Barnstable counties with 0.0,0.0, 0.1, 0.7, and 1.0 percent of Massachusetts spending, respectively. 5.1 County-Level Empirical Analysis We run a triple difference-in-differences estimation by interacting the 2005 insurance level with Massachusetts-specific time dummies. Kolstad and Kowalski [2012] and Miller [2012] show that, prior to 2006, the uninsured rate in Massachusetts was fairly steady, and thenin2007and2008ittookasteepfall. Thisimpliesthattherewasanexogenousincrease in insurance coverage after 2005 attributable to the reform. The difference-in-differences regression takes the form: 38 To keep the table smaller we do not depict statistics for Dukes, Nantucket, Franklin, Hampshire, and Barnstable counties which together represent 7 percent of Massachusetts s population. All counties were used in our estimation. 19

21 ln(p jnct ) = t=2010:q4 t=2003:q2 [λ t Uninsured2005 c 1(MA) 1(Time t ) + γ 1,t 1(MA) 1(Time t )+γ 2,t Uninsured2005 c 1(Time t )+δ t 1(Time t )] + [α jc 1(Service jc )]+ [α s 1(State s ) Time t ]+X n β +Y ct θ +ε jnct,(2) jc s where n represents the patient receiving treatment, c represents the provider s county, and j represents three characteristics of the service CPT code plus modifier (k), provider ID(i), andhealthinsurancetype(l). ItfollowsthatP jnct isthepriceofservice j paidtoaprovider who resides in county c and consumed by patient n in year-quarter t. Including service fixed effects, 1(Service jc ), isolates variationover time in the payment to a specific provider, procedure and type of plan (that is, HMO, PPO, POS, etc.). We include the county, c, as a characteristic in the fixed effect to account for the fact that the provider may move geographic areas. The variable 1(MA) is a dummy variable representing a service provided by a physician located in Massachusetts. The variable labeled Uninsured2005 c measures the 2005 uninsured rate in county c shown in Table 3 for Massachusetts. We weight observations in the regression by a proxy for the service s relative value units (RVUs). Our proxy is the average price of the CPT code over the entire U.S. sample the same as that used in Dunn, Shapiro, and Liebman [2012]. We show that our main result does not depend on weighting. Standard errors are clustered by county to account for spatial correlation between counties and time-series correlation in the error terms. 39 The specification includes a number of controls. The 2005 uninsured level variables are interacted with the time dummies, Uninsured2005 c 1(Time t ), to control for any timevarying unobservable characteristics attributable to the 2005 uninsured rate level in the county. The vector Y ct represents the county-level unemployment rate and median income in period t, and the vector X n represents demographic variables including patient n s age and gender. Importantly, the coefficients on these control variables are estimated using variation in the entire sample prices in the states other than Massachusetts. Estimating specification (2) on the New England states implies that we are removing any time-varying component attributable to the 2005 uninsured level or income level, that is common to all counties in New England. 40 Finally, the Massachusetts time dummies, 1(MA) 1(Time t ), 39 For robustness purposes, we also clustered at the state level which produced smaller standard errors. This indicates that errors are spatially correlated at the county level, but there is likely negative correlation in the errors between counties. 40 Estimates on the sample of Northeast states are available in the appendix. 20

22 control for time-varying unobservable factors common to all counties in Massachusetts, while the main time dummies, 1(Time t ), control for time-varying unobservable variables common to all states. Since these main time dummies control only for aggregate time factors, we include state-specific time trends, 1(State s ) Time t, in our specification. 41 Pertinent tothisanalysisaretheλ t estimates the coefficientsontheinteractionsofthe 2005 uninsured rate in county c, time dummies, and a Massachusetts time dummy. These coefficients track the difference in the pattern of prices over time within Massachusetts that are attributable to different 2005 uninsured rate levels in the county. More specifically, since each 1 percentage point difference in the 2005 uninsured rate level translates into a 0.50 to 0.75 expected percentage point increase in coverage, the λ t coefficients can be interpreted as the impact of an expected 0.50 to 0.75 percentage point increase in insurance coverage on the log price level in period t relative to the base period, 2003q1. Estimates of λ t are shown in Figure 3. To track how these estimates align with the sequence of events leading up to the Massachusetts reform, we depict four major event indicators as vertical lines. The first line indicates the beginning of the pre-reform period, 2004q4, when Romney published The Boston Globe, editorial. The other three lines indicate when Romney publicly endorsed the individual insurance mandate (2005q2), the date health insurance reform passed the Massachusetts House and Senate separately (2005q4), and the date the legislation was finally enacted (2006q2). The pattern of the λ t s shows a a clear level shift that occurs in the first quarter of 2006 the quarter immediately after the passage of the legislation in the House and Senate, but before the bill s final enactment in the second quarter of 2006 (April). This pattern indicates that prices were likely negotiated up in anticipation of the reform eventually being enacted. Given Romney s previous public support of the bill, it is certainly plausible that insurers set payment rates in such an anticipatory fashion. In the appendix, we perform a few robustness exercises. First, we estimate our model on the sample of Northeast states, which includes the sample of New England states plus New York, New Jersey, Maryland, Pennsylvania, and Delaware. Second, we include trends specific to the 2005 uninsured rate in the Massachusetts counties. All of these specifications also show a similar level shift in prices. 41 The inclusion of state-time trends have proven critical in some difference-in-differences analysis, as demonstrated by Besley and Burgess [2004] looking at the effects of labor regulations on businesses in Indian states. Our results are similar without state-time trends, although their inclusion appears to reduce our standard errors slightly. 21

23 Figure 3: Estimates of λ t Globe Editorial Romney endorses individual mandate Passes House & Senate Legislation Enacted Notes: Estimates and 95-percent confidence intervals using the New England samples of the MarketScan data. Each diamond represents the coefficient λ t in specification (2). Standard errors are clustered by county. Coefficients are in column 1 of the table in Appendix B. 5.2 Measuring the Implied Impact of the Reform To estimatethetimingandsizeofthelevel shift inλ t, weestimateaseries ofregressions that take the following form: ln(p jnct ) = λ t Uninsured2005 c 1(MA) 1(Time t t ) + t=2010:q4 t=2003:q2 [γ 1,t 1(MA) 1(Time t )+γ 2,t Uninsured2005 c 1(Time t )+δ t 1(Time t )] + [α jc 1(Service jc )]+ [α s 1(State s ) Time t ]+X n β +Y ct θ+ε jnct. (3) jc s This specification replaces the flexibly estimated coefficients of λ t with the post-period coefficient λ t. Specifically, λ t can be interpreted as the impact of an expected 0.50 to 0.75 percentage point increase in insurance coverage on the log price level in the post-period, defined as t t, relative to the pre-period, defined as t < t. Note that if the estimate λ t is statistically significantly different than zero, we can reject the null hypothesis that there is no break in the level of λ t in period t. 22

24 Figure 4: Estimates of λ t t* Notes: Estimates and 95-percent confidence intervals using the New England MarketScan data. Each diamondrepresentsthecoefficientλ t estimatedfromasingleregressionofspecification(3). Theregression is run 21 times, once for each t [2003q3,2008q3]. Standard errors are clustered by county. Coefficients are in column 1 of Table 4. The date of the reform is marked by the red line. We run the specification 21 times, once for each t [2003q3,2008q3], then collect the corresponding λ t s and standard errors. Estimates run on the New England sample are plotted in Figure 4. The estimate of λ t is statistically insignificant at the 5 percent level until t reaches 2005q4. The estimate reaches a global maximum of in 2006q1 with a corresponding p-value of 0.009, and then becomes statistically insignificant. This exercise shows that of all the time periods between 2003q3 and 2008q3, the shift in pricing, captured by λ t, most likely occurred in 2006q1. This break in λ t occurs over a year after Romney s editorial, but one quarter bef ore the insurance reform legislation was enacted in 2006q2. Specifically, this period corresponds to the period between when the legislation passed the Massachusetts House and Senate separately and the actual enactment of the reform. We investigated the robustness of this result by repeating this exercise under a number of alternative specifications. Table 4 shows the estimates and p-values of λ t under our main specification (the first two columns) as well as four alternatives. To address concern that our provider ID threshold (being in the sample at least six years) is either too strict or too narrow, we ran a specification where the provider ID is in the sample all eight years of the sample and another specification where there is no restriction. Another concern is 23

25 Table 4: Robustness: Estimates of λ t t λ t p-value λ t p-value λ t p-value λ t p-value λ t p-value λ t p-value 2003q q q q q q q q q q E q E q E q q q q q q q q q Provider ID in sample at least 6 years X X X X Provider ID in sample all 8 years X No Provider ID Restrictions X Weighting (by RVU) X X X X X State Time Trends X X X X X MA County-Uninsured-Rate Time Trends X County-Provider-CPT-Plan Fixed Effects X X X X X County-Provider-CPT Fixed Effects X Plan Fixed Effects X Notes: Estimates and p-values of estimates using the New England MarketScan data. The first column repeats estimates of λ t shown in Figure 4 where the regression from equation (3) is run 21 times, once for each t [2003q3,2008q3]. The subsequent columns repeat this exercise for different samples and specifications. Estimates with the smallest p-value are marked in bold face. 24

26 that including the type of health plan, in addition to the provider ID and CPT code, in the fixed effect may be an overly strict assumption. We relax this assumption by including separate plan-type fixed effects. In the final two columns, we remove state-time trends and observation weights from the regression. Finally, there may be a concern the result is due to trends specific to counties within Massachusetts-specific uninsured-rate trends. To address this we include Massachusetts count uninsured-rate time trends. In all of these specifications, the largest estimate of λ t and smallest p-value are obtained when t is set to 2006q1. We can use the estimate of λ t =2006q1 to measure the effect of an expected increase in insurance coverage attributable to the reform. Note that, if price setters are forward looking, the estimate can be interpreted as the impact of an expected 0.50 to 0.75 percentage point increase in insurance coverage attributable to the reform. Multiplying the estimate by (1/0.75) therefore translates the estimate into the lower-bound impact of a 1 percent expected increase in coverage. To measure the expected impact of the entire reform, we would then further multiply this figure by the total number of percentage points the reform was expected to increase the insurance coverage rate. We use the 4.5 percentage point estimate from the DHCFP. It follows that: Impact of Expected Insurance Increase = [exp((1/0.75) 4.5 λ t =2006q1) 1] = 10.8 percent (4) where λ t =2006q1 = Because overall prices in Massachusetts grew 69 percent (42 percentage points above the U.S. average) between 2003 and 2010, this estimate means that approximately one-sixth of the price growth in Massachusetts between 2003 and 2010 is directly attributable to the expected insurance increase of the reform. 6 Conclusion Our study confirms that there was likely a reaction in physician prices in anticipation of health insurance reform in Massachusetts. Service prices rose by at least 10.8 percent, mainly in the period after the reform passed the House and Senate separately. This is consistent with insurers and physicians negotiating prices in a forward-looking manner, in anticipation of the implementation of the reform. This study does not explicitly identify the mechanism by which prices were raised. That is, it is not clear whether demand or competitive concerns caused prices to rise in 25

27 anticipation of the reform. However, the available data from the DHCFP suggest that competitive concerns in the insurance market may have played an important role. The DHCFP reports that premiums grew less rapidly than expenditures, while the medical loss ratio increased. In addition, despite our finding of a price increase, Graves and Gruber [2012] find that the reforms had no effect on group premiums in Massachusetts, and even led to lower permiums in the nongroup market, relative to other states. This evidence suggests that service price increases were not passed on to consumers, which is consistent with a more competitive insurance market post-reform. Furthermore, our conversations with Massachusetts health care providers and insurers leads us to believe that there was at least a perceived increase in the degree of competition in the insurance market stemming from the exchange. In terms of implications for the ACA, our results imply that there may be price effects and they could occur before implementation. A major caveat, however, is that there are important differences between the ACA and the Massachusetts reform in terms of implementation. The Massachusetts Connector opened almost immediately after enactment, in October Similarly, the individual mandate in Massachusetts went into effect for the 2006 tax year. By contrast, the ACA implies that the insurance exchanges and individual mandate take effect in October 2013 and January 2014, respectively years after the March 2010 enactment. It is not clear how a policy lag of this length would have affected anticipation and the results in this study. While this study highlights an important consequence of the reform in Massachusetts, the results should not be treated as a normative statement regarding the merits of the reform. To fully evaluate the success of a reform, policymakers must weigh the benefits of the reform, such as expanded coverage and additional competition in the insurance market, against the associated cost. The analysis presented in this paper provides information with regards to medical-care prices, allowing for more accurately shaped policies in this regard. References [1] Abadie, A., Diamond, A., and J. Hainmueller Synthetic Control Methods for Comparative Case Studies: Estimating the Effect of California s Tobacco Control Program. Journal of the American Statistical Association [2] Bebinger, M Personal Responsibility: How Mitt Romney Embraced The Individual Mandate In Massachusetts Health Reform. Health Affairs. 31(9):

28 [3] Brown, J. and A. Goolsbee Does the Internet Make Markets More Competitive? Evidence from the Life Insurance Industry Journal of Political Economy. 110(3) [4] Besley, T. and R. Burgess Can Labour Market Regulation hinder Economic Performance? Evidence from India. Quarterly Journal of Economics 113, [5] Buchmueller, T., DiNardo, J., and R. Valletta The Effect of an Employer Health Insurance Mandate on Health Insurance Coverage and the Demand for Labor: Evidence from Hawaii. American Economic Journal: Economic Policy. [6] Clemens, J. and J. Gottlieb Bargaining in the Shadow of a Giant: Medicare s Influence on Private Payment Systems. NBER Working Paper. [7] Clemens, J. and J. Gottlieb Do Physicians Financial Incetives Affect Treatment Patterns and Patient Health? American Economic Review, Forthcoming. [8] Dafny, L., M. Duggan, and S. Ramanarayanan Paying a Premium on Your Premium? Consolidation in the U.S. Health Insurance Industry, American Economic Review, 102(2), pp [9] Division of Health Care Finance and Policy. Health Care in Massachusetts: Key Indicators. May 2011 Edition. [10] Deprez, R., Glied, S., Kappel, S., Louie, M., Perry, B., Robertson, B., Rodriguez, K., Schwabe, N., and Tilipman, N Achieving Universal Coverage through Comprehensive Health Reform: The Vermont Experience. Report of Findings. Portland, ME: University of New England, Center for Health Policy, Planning, and Research. [11] Dunn, A. and A.H. Shapiro Do Physicians Possess Market Power? The Journal of Law and Economics, forthcoming. [12] Dunn, A., A.H. Shapiro, and E. Liebman Geographic Variation in Commercial Medical Care Expenditures: A Decomposition Between Price and Utilization, Journal of Health Economics. [13] Ellison, G., and S.F. Ellison Strategic Entry Deterrence and the Behavior of Pharmaceutical Incumbents Prior to Patent Expiration. American Economic Journal: Microeconomics 27

29 [14] Ericson, K.M. and A. Starc Heuristics and Heterogeneity in Health Insurance Exchanges: Evidence from the Massachusetts Connector. American Economic Review: Papers & Proceedings. 102(3): [15] Finkelstein, A., (2007), The Aggregate Effects of Health Insurance: Evidence from the Introduction of Medicare, Quarterly Journal of Economics, 122(3) pgs [16] Finkelstein, A., S. Taubman, B. Wright, M. Bernstein, J. Gruber, J. Newhouse, H. Allen, K. Baicker, (2012), The Oregon Health Insurance Experiment: Evidence From the First Year, Quarterly Journal of Economics, 127(3) pgs [17] Gali, J. and M. Gertler Inflation Dynamics: A structural econometric analysis. Journal of Monetary Economics. [18] Goetz, C. and A.H. Shapiro Strategic Alliance as a Response to the Threat of Entry: Evidence from Airline Codesharing. International Jouranl of Industrial Organization. [19] Goolsbee, A. and C. Syverson How Do Incumbents Respond to the Threat of Entry? Evidence from the Major Airlines. Quarterly Journal of Economics. [20] Gruber, J The Impacts of the Affordable Care Act: How Reasonable are the Projections? National Tax Journal. [21] Kolstad, J. and Kowalski, A The Impact of Health Care Reform on Hospital and Preventive Care: Evidence from Massachusetts. The Journal of Public Economics. Vol [22] Krueger, A. and Kuziemko, I The Demand for Health Insurance Among Uninsured Americans: Results of a survey experiement and implications for policy. Journal of Health Economics. Vol [23] Shumilkina, E. and J. Kwoka The Price Effect of Eliminating Potential Competition: Evidence from An Airline Merger. Journal of Industrial Economics. [24] Long, S What Is the Evidence on Health Reform in Massachusetts and How Might the Lessons from Massachusetts Apply to National Health Reform?. Urban Institute Memo. [25] Long, S., and Stockley, K Sustaining Health Reform In A Recession: An Update On Massachusetts As Of Fall Health Affairs. 28

30 [26] Long, S., Stockley, K. and H. Dahlen Massachusetts Health Reforms: Uninsurance Remains Low, Self-Reported Health Status Improves as State Prepares to Tackle Costs. Health Affairs. 29(6): [27] Long, S., Stockley, K. and A. Yemane Another Look at the Impacts of Health Reform in Massachusetts: Evidence Using New Data and a Stronger Model. American Economic Review: Papers & Proceedings. 99(2): [28] Moulton, B An Illustration of a Pitfall in Estimating the Effects of Aggregate Variables on Micro Units. Review of Economics and Statistics. [29] McDonough, J., Rosman, B., Butt, M., Tucker, L., and Howe, L.K Massachusetts Health Reform Implementation: Major Progress and Future Challenges. Health Affairs, 27, no [30] Manning, W., J. Newhouse, N. Duan, E. Keeler, and A. Leibowitz, (1987), Health Insurance and the Demand for Medical Care: Evidence from a Randomized Experiment, American Economic Review, 77(3) pgs [31] Miller, S. 2012a. The Effect of Insurance on Emergency Room Visits: An Analysis of the 2006 Massachusetts health reform Journal of Public Economics. [32] Miller, S. 2012b. The Effect of the Massachusetts Reform on Health Care Utilization Inquiry, 49(4) pgs [33] Reinhardt, Uwe The Pricing of U.S. Hospital Services: Chaos Behind A Veil of Secrecy Health Affairs, 25(1) pgs [34] Tenn, S. and B. Wendling Entry Threats and Pricing in the Generic Drug Industry. Review of Economics and Statistics 29

31 A Synthetic Control Analysis Table 5: Synthetic Massachusetts State Weight State Weight State Weight State Weight Alabama 0 Montana 0 Illinois South Carolina 0 Alaska Nebraska 0 Indiana 0 South Dakota 0 Arizona 0 Nevada 0 Iowa 0 Tennessee 0 Arkansas 0 New Hampshire Kansas 0 Texas 0 California 0 New Jersey 0 Kentucky 0 Utah 0 Colorado 0 New Mexico 0 Louisiana 0 Vermont 0 Connecticut 0 New York 0 Maine 0 Virginia 0 Delaware 0 North Carolina 0 Maryland 0 Washington 0 DC 0 North Dakota 0 Michigan 0 West Virginia 0 Florida 0 Ohio 0 Minnesota 0 Wisconsin 0 Georgia 0 Oklahoma 0 Mississipppi 0 Wyoming 0 Hawaii 0 Oregon 0 Missouri 0 Idaho 0 Pennsylvania 0 Rhode Island Table 6: Massachusetts Predictor Means Variables Real Synthetic Percent of Population Over Uninsurance Rate Median Household Income ($) Unemployment Rate Percent of Spending HMO Percent of Spending PPO University Hospitals per Capita 7.04E E-06 Median House Value Median Rent Perc. of Pop. with College Degree Population Density (persons per sq. mile) γ 2004: γ 2005: γ 2006: Notes: All variables except lagged γ t are averaged for the 2003q1 to 2006q1 period. PPO and HMO spending are calculated from MarketScan. Other variables we obtained from the Area Resource File.

32 Figure 5: Physician Service Price Growth by State: Relative to Synthetic Control. (No pre-reform γ t in predictor set) 14 Total Price Growth: Full Sample (2003q1 to 2010q4) 14 Annualized Price Growth: Pre-Reform (2003q1 to 2004q4) Frequency 8 6 Frequency MA 2 MA Annualized Price Growth: Lead-up to Reform (2004q4 to 2006q2) 14 Annualized Price Growth: Post-Reform (2006q2 to 2010q4) Frequency 8 6 Frequency MA 2 MA 0 0 Notes: Histogram of physician service price growth by state relative to each state s synthetic control. For example, the cumulative growth between 2003q1 and 2010q4 is exp(γ gap 2010q4 ) 1. Massachusetts value in the histogram is marked. 31

33 B Robustness: Estimates of λ t We perform a few robustness exercises. First, we estimate our model on the sample of Northeast states, which includes the sample of New England states plus New York, New Jersey, Maryland, Pennsylvania, and Delaware. This sample includes 37,493,340 observations. Second, we include trends specific to the 2005 uninsured rate in the county (i.e. trend Uninsured2005 c ) as well as specific to Massachusetts (i.e. trend Uninsured2005 c 1(MA). We perform two specifications that include this type of trend. One specification estimates this trend using the entire sample period, the other specification estimates the trend on the pre-reform period. 32

34 New England Sample Northeast Sample 2003q * ** ** (0.002) (0.002) (0.002) (0.002) (0.002) (0.002) 2003q ** *** *** (0.003) (0.003) (0.003) (0.002) (0.002) (0.002) 2003q (0.005) (0.004) (0.004) (0.004) (0.003) (0.003) 2004q (0.010) (0.009) (0.009) (0.008) (0.006) (0.006) 2004q (0.013) (0.011) (0.012) (0.011) (0.009) (0.009) 2004q (0.011) (0.009) (0.009) (0.009) (0.007) (0.006) 2004q (0.010) (0.007) (0.007) (0.008) (0.006) (0.005) 2005q (0.007) (0.004) (0.003) (0.006) (0.003) (0.002) 2005q ** *** * (0.009) (0.005) (0.002) (0.007) (0.003) (0.002) 2005q ** *** * (0.009) (0.005) (0.002) (0.007) (0.003) (0.002) 2005q * (0.009) (0.005) (0.007) (0.002) 2006q * 0.011* 0.017*** 0.022** 0.010* 0.009** (0.011) (0.006) (0.006) (0.010) (0.005) (0.004) 2006q ** 0.020** (0.011) (0.005) (0.006) (0.010) (0.004) (0.003) 2006q ** 0.022** * (0.011) (0.005) (0.006) (0.010) (0.005) (0.004) 2006q * 0.010* 0.018*** 0.022** (0.012) (0.005) (0.006) (0.010) (0.004) (0.004) 2007q ** (0.010) (0.003) (0.006) (0.008) (0.002) (0.003) 2007q * 0.021*** 0.004** (0.010) (0.003) (0.006) (0.008) (0.002) (0.004) 2007q ** (0.011) (0.003) (0.007) (0.008) (0.002) (0.004) 2007q *** (0.011) (0.002) (0.008) (0.008) (0.002) (0.005) 2008q ** 0.020** 0.029** 0.008** (0.014) (0.004) (0.008) (0.012) (0.004) (0.004) 2008q * 0.010** 0.021** 0.029** 0.008** (0.014) (0.004) (0.009) (0.012) (0.003) (0.004) 2008q ** 0.019** 0.028** 0.006** (0.014) (0.003) (0.009) (0.011) (0.003) (0.004) 2008q * 0.008** 0.020** 0.030** 0.006** (0.014) (0.003) (0.010) (0.012) (0.003) (0.005) 2009q ** (0.014) (0.003) (0.009) (0.011) (0.002) (0.005) 2009q * 0.018* 0.029** 0.003* (0.015) (0.002) (0.009) (0.012) (0.002) (0.005) 2009q ** 0.019* 0.030** (0.015) (0.002) (0.010) (0.012) (0.002) (0.006) 2009q * 0.030** (0.015) (0.002) (0.011) (0.012) (0.002) (0.006) 2010q ** (0.015) (0.002) (0.011) (0.013) (0.001) (0.006) 2010q ** 0.003** (0.016) (0.002) (0.011) (0.014) (0.001) (0.006) 2010q ** (0.016) (0.002) (0.012) (0.014) (0.001) (0.006) 2010q ** (0.016) (0.013) (0.014) (0.007) trend*uninsured (0.000) (0.001) (0.000) (0.000) trend*uninsured2005*ma ** 0.001* (0.001) (0.001) (0.000) (0.001) Full Sample Period Uninsurance Trends X X Pre-2006 Uninsurance Trends X X